Agentic Project Management

Coding agents now carry out substantial software tasks with growing autonomy. As a task grows, the agent must hold the project’s intent while accumulating context competes for the model’s finite attention. Recent solutions address this by recording intent in persistent documents and by dividing work among multiple agents.

We present Agentic Project Management (APM), an open-source framework that operates on top of interactive coding agents. APM combines persistent planning documents with role-specific division of work, and makes its coordination user-mediated: its agents communicate through the developer, allowing them to review the work behind each message and request corrections before passing it on. How the work is divided and assigned sets how often such opportunities arise. The design aims to keep authority and decision-making over development with the user.

To evaluate it, we propose a reproducible methodology for comparing frameworks on top of coding agents and similar human-in-the-loop workflows. We implement it as a within-participant case study of APM against GitHub’s Spec-kit, where twelve developers each completed the same demanding task across two sessions, once with each framework, on Claude Code with Claude Opus 4.7 or Claude Sonnet 4.6. The comparison covers economic cost, each session’s temporal and interaction shape, functional test scores, and developer perception. APM cost more in nine of the twelve pairs, adding a median of $6.10 across the cohort, with the higher cost concentrated under Opus, and took longer in all twelve pairs, adding a median of 36 minutes to session duration. The frameworks carried no firm difference in functional outcome across the cohort, although APM showed a more favourable pattern under Sonnet. Even so, eleven of the twelve participants rated APM higher on effectiveness and nine on confidence. Our study’s comparison documents the added time and cost of APM’s user-mediated coordination alongside its opportunities for developer control and review.

For attribution, please cite this work as:

K. Chaidemenos, “Agentic Project Management: A Spec-Driven Multi-Agent Framework for AI-Assisted Software Engineering,” BSc thesis, Department of Informatics and Telecommunications, National and Kapodistrian University of Athens, Athens, Greece, Oct. 2026.

Or with BibTeX:

@thesis{chaidemenos2026apm,
    author = {Chaidemenos, Konstantinos},
    title = {{Agentic Project Management: A Spec-Driven Multi-Agent Framework for AI-Assisted Software Engineering}},
    type = {BSc thesis},
    institution = {{Department of Informatics and Telecommunications, National and Kapodistrian University of Athens}},
    location = {Athens, Greece},
    month = oct,
    year = {2026},
}